linux-nvme.lists.infradead.org archive mirror
 help / color / mirror / Atom feed
From: Sagi Grimberg <sagi@grimberg.me>
To: Bart Van Assche <bvanassche@acm.org>, Christoph Hellwig <hch@lst.de>
Cc: Keith Busch <kbusch@kernel.org>,
	linux-nvme@lists.infradead.org, Hannes Reinecke <hare@suse.de>,
	Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Subject: Re: [PATCH] nvmet-rdma: Release connections synchronously
Date: Mon, 22 May 2023 10:07:53 +0300	[thread overview]
Message-ID: <9abb3c36-9b1c-75c6-b071-1364c30dc84b@grimberg.me> (raw)
In-Reply-To: <11382811-caa7-c64c-5fe5-659fc5722c44@grimberg.me>


>>>> @@ -1582,11 +1566,6 @@ static int nvmet_rdma_queue_connect(struct 
>>>> rdma_cm_id *cm_id,
>>>>           goto put_device;
>>>>       }
>>>> -    if (queue->host_qid == 0) {
>>>> -        /* Let inflight controller teardown complete */
>>>> -        flush_workqueue(nvmet_wq);
>>>> -    }
>>>> -
>>>>       ret = nvmet_rdma_cm_accept(cm_id, queue, &event->param.conn);
>>>>       if (ret) {
>>>>           /*
>>>
>>> You could have simply removed this hunk alone to make lockdep quiet on
>>> this, without the need to rework the async queue removal.
>>>
>>> The flush here was added to prevent a reset/connect/disconnect storm
>>> causing the target to run out of resources (which we have seen reports
>>> about in the distant past). What prevents it now?
>>>
>>> And you both reworked the teardown, and still removed the flush, I don't
>>> get why both are needed.
>>
>> Hi Sagi,
>>
>> My understanding is that the above flush_workqueue() call waits for 
>> prior release_work to complete. If the release_work instance is 
>> removed, I don't think that the above flush_workqueue() call is still 
>> necessary.
> 
> I'm wandering if making delete_ctrl synchronous may be a problem
> in some cases, nvmet_fatal_error can be triggered from the
> rdma completion path (rdma core workqueue context). Wondering if there
> may be some inter-dependency there...

For example, nvmet_rdma_comp_error -> nvmet_fatal_error -> delete_ctrl
-> __nvmet_rdma_queue_disconnect -> nvmet_rdma_free_queue ->
nvmet_rdma_destroy_queue_ib -> ib_drain_qp

ib_drain_qp cannot complete because the completion handler is
synchronously deleting the queue, hence cannot reap the drain
completion...


      reply	other threads:[~2023-05-22  7:08 UTC|newest]

Thread overview: 8+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-05-11 15:03 [PATCH] nvmet-rdma: Release connections synchronously Bart Van Assche
2023-05-14 10:38 ` Shinichiro Kawasaki
2023-05-15 17:29   ` Bart Van Assche
2023-05-29 11:49     ` Pawel Baldysiak
2023-05-17  7:42 ` Sagi Grimberg
2023-05-18 20:32   ` Bart Van Assche
2023-05-22  6:41     ` Sagi Grimberg
2023-05-22  7:07       ` Sagi Grimberg [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=9abb3c36-9b1c-75c6-b071-1364c30dc84b@grimberg.me \
    --to=sagi@grimberg.me \
    --cc=bvanassche@acm.org \
    --cc=hare@suse.de \
    --cc=hch@lst.de \
    --cc=kbusch@kernel.org \
    --cc=linux-nvme@lists.infradead.org \
    --cc=shinichiro.kawasaki@wdc.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).